Human-Centric AI in Complex Scenarios

Artificial intelligence promises to augment people by combining two goals: automation and collaboration with people. Human-AI partnerships can make a significant impact on complex tasks that are more ambiguously structured, effectively or actually unbounded, involve large amounts of data, and cannot be solved analytically in polynomial time. Its realization in practice is in an early stage, as the strengths and weaknesses of AI are still heavily studied, model development is in flux, and AI may lack explainable, collaborative, adaptive, and responsible teaming mechanisms. This chapter considers the need, current approaches, and standing challenges of human-centric AI in four critical domains with many complex tasks: content safety, education, traffic understanding, and robotics. In the domain of content safety, we review state-of-the-art commonsense technology that can facilitate a deep understanding of malicious undertones in internet memes and fallacious arguments, to deal with phenomena of hate speech and misinformation. In education, we describe methods for personalizing materials to users through analogical recommendations and tutoring systems that can guide human understanding over time. In traffic monitoring, we review complex reasoning tasks, such as counterfactual reasoning, to understand the causes, implications, and possible ways to avoid traffic events. In robotics, we describe a recent framework for robot manipulation tasks that consolidates visual reasoning, object manipulation, and constraint satisfaction. We conclude the chapter with a summary of the lessons learned, standing challenges, and possible future directions for these and other domains.